x
It should answer the questions it can answer, route the ones it can't, and be honest about which is which. Almost none of them do.
The standard version — "I'm away until the 11th and will respond to your message on my return" — is a promise most of us quietly break. You come back to 200 emails, triage the loudest, and the person who wrote politely on day two gets a reply on day fourteen or never. The auto-reply didn't hold anything. It just delayed the disappointment.
The fix isn't a cleverer message. It's deciding, before you go, which inbound your business can genuinely handle without you — and building cover for exactly that.
Sort your inbound into three tiers. Most one-person businesses find nearly everything falls into the first two.
Tier 1 — answerable from what you've already written. Pricing, turnaround, what you do and don't take on, how to book, where to send a file. You have answered these hundreds of times. The answers are stable, and they're already sitting in your sent folder.
Tier 2 — routable. Someone needs a thing that exists but isn't a conversation: an invoice copy, a booking link, a document you always send at that stage. No judgement required, just retrieval.
Tier 3 — needs you. A new brief, a complaint, anything with money or a deadline attached that isn't already agreed. This tier must never be automated. It gets acknowledged honestly and held.
Cover means Tier 1 and Tier 2 get a real answer within minutes, and Tier 3 gets a message that doesn't lie. Something closer to: "I'm away until the 11th. If this is a new project or something time-sensitive, I'll pick it up on the 11th and reply that day — I'm not monitoring in between." That's a promise you can keep, because it's smaller.
Building the Tier 1 layer takes an afternoon. Pull your last fifty replies, find the twenty questions you answer repeatedly, and write each answer once, properly. That library is the asset — the tool you point at it barely matters, which is the same argument behind the one-pager that doubles every AI answer. Context beats configuration nearly every time.
Three things, and they're the same three whether you're away for a week or building this permanently.
It must never quote a price it wasn't given. It must never agree to a deadline. It must never claim something is done. Every one of those creates an obligation you didn't sign, and an automated system will produce them fluently because fluency is what it's for — which is why it's worth making it show its working on anything you'll be held to.
Write the escalation rule down before you switch anything on: what does this system do when it doesn't know? The honest answer — "I don't have that, and it needs a person; you'll hear on the 11th" — is a better outcome than a confident guess, and it's the single boundary that makes the whole arrangement safe. That decision deserves its own twenty minutes, which is why the five-check trust test puts it before anything else.
Here's the checkable bit, and it lands mid-holiday season.
Under the EU AI Act, transparency obligations — including telling people when they're dealing with an AI system — apply from 2 August 2026. Machine-readable marking of AI-generated content follows for systems already on the market from 2 December 2026, with a simplified route for smaller businesses.
Practically: if your holiday cover replies to customers, say so. One line — "This is an automated reply" — at the top. It costs nothing, it's the honest thing regardless of jurisdiction, and it stops the moment where someone realises mid-thread they've been talking to software and reconsiders everything they've been told. Disclosure isn't a tax on automation. It's what makes automation survivable in a business built on trust. (Not legal advice — check your own obligations.)
Most people discover this the awkward way: a customer asks "is this a real person?" and the answer arrives too late to be reassuring.
Don't build the system first. Build the answer library first — twenty questions, twenty honest answers, written once. Everything downstream is plumbing, and plumbing is cheap.
If you want a shortlist of tools that do this without a monthly subscription per seat, the free 5-min read is here: 10 AI Tools That Save 10 Hours a Week.
Related Decodes
TL;DR: A useful out-of-office answers the questions it can, routes the ones it can't, and honestly holds the rest — instead of promising a reply you won't send. Sort inbound into answerable, routable and needs-a-human, automate only the first two, and write the escalation rule before switching anything on. From 2 August 2026 the EU AI Act requires telling people when they're dealing with an AI system, so add the disclosure line at the top.
